veyselozdemir/nnUNet-Breast-Cancer-Ultrasound

🎗️ nnU-Net Model for Breast Ultrasound Segmentation

This repository contains a trained nnU-Net v2 model for breast ultrasound segmentation. The model was trained using a custom dataset following the nnU-Net v2 pipeline and is now available for inference and further fine-tuning. For more information about the dataset, visit the repository.

📌 Model Details

• Framework: nnU-Net v2
• Task: 2D Medical Image Segmentation
• Dataset: Breast Ultrasound Dataset (Dataset ID: 101)
• Training Folds: All folds
• Training Epochs: 10 (nnUNetTrainer_10epochs)
• Trainer Configuration: 2d
• Checkpoints: checkpoint_final.pth

---

📁 Folder Structure

nnUNet_results/
└── Dataset101_Breast/
    └── nnUNetTrainer_10epochs__nnUNetPlans__2d/
        ├── fold_all/
        │   └── checkpoint_final.pth  # Trained model weights
        ├── dataset.json  # Trained dataset metadata
        └── plans.json  # Preprocessing and training plans

---

🚀 How to Use This Model

1. Ensure you have nnU-Net v2 installed and configured all the essentials to run the framework. For more information, see the documentation.
2. Add the nnUNet_results folder to the root of your project as it is.
3. Open the terminal, ensure that the directory is at the root of the project, and run the following inference command.

   nnUNetv2_predict -i path/to/your/images \
                 -o path/to/your/output/folder \
                 -d 101 -c 2d -f all -tr nnUNetTrainer_10epochs
   

  ⚠️ If you run this on a Mac, add -device mps in the inference command to leverage the GPU.

---

👥 Contact

For questions or collaboration, reach out at:

• Email: ozdemirsoftware.dev@gmail.com
• GitHub: veysel-ozdemir